Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Latent Space Score-based Diffusion Model for Probabilistic Multivariate Time Series Imputation

Liang, Guojun ; Abiri, Najmeh LU ; Hashemi, Atiye Sadat LU ; Lundström, Jens ; Byttner, Stefan and Tiwari, Prayag (2025)
Abstract
Accurate imputation is essential for the reliability
and success of downstream tasks. Recently, diffusion models have
attracted great attention in this field. However, these models neglect the latent distribution in a lower-dimensional space derived
from the observed data, which limits the generative capacity
of the diffusion model. Additionally, dealing with the original
missing data without labels becomes particularly problematic.
To address these issues, we propose the Latent Space ScoreBased Diffusion Model (LSSDM) for probabilistic multivariate
time series imputation. Observed values are projected onto
low-dimensional latent space and coarse values of the missing
data are reconstructed without... (More)
Accurate imputation is essential for the reliability
and success of downstream tasks. Recently, diffusion models have
attracted great attention in this field. However, these models neglect the latent distribution in a lower-dimensional space derived
from the observed data, which limits the generative capacity
of the diffusion model. Additionally, dealing with the original
missing data without labels becomes particularly problematic.
To address these issues, we propose the Latent Space ScoreBased Diffusion Model (LSSDM) for probabilistic multivariate
time series imputation. Observed values are projected onto
low-dimensional latent space and coarse values of the missing
data are reconstructed without knowing their ground truth
values by this unsupervised learning approach. Finally, the
reconstructed values are fed into a conditional diffusion model
to obtain the precise imputed values of the time series. In
this way, LSSDM not only possesses the power to identify the
latent distribution but also seamlessly integrates the diffusion
model to obtain the high-fidelity imputed values and assess the
uncertainty of the dataset. Experimental results demonstrate
that LSSDM achieves superior imputation performance while
also providing a better explanation and uncertainty analysis
of the imputation mechanism. The website of the code is
https://github.com/gorgen2020/LSSDM imputation.
Index Terms—Diffusion model, multivariate time series, imputation, variational graph autoencoder (Less)
Please use this url to cite or link to this publication:
author
; ; ; ; and
publishing date
type
Contribution to conference
publication status
published
DOI
10.48550/arXiv.2409.08917
language
English
LU publication?
no
id
75a4b991-3c74-4e79-9262-a9fc52e4cfc7
date added to LUP
2025-01-31 14:32:59
date last changed
2025-02-03 10:21:31
@misc{75a4b991-3c74-4e79-9262-a9fc52e4cfc7,
  abstract     = {{Accurate imputation is essential for the reliability<br/>and success of downstream tasks. Recently, diffusion models have<br/>attracted great attention in this field. However, these models neglect the latent distribution in a lower-dimensional space derived<br/>from the observed data, which limits the generative capacity<br/>of the diffusion model. Additionally, dealing with the original<br/>missing data without labels becomes particularly problematic.<br/>To address these issues, we propose the Latent Space ScoreBased Diffusion Model (LSSDM) for probabilistic multivariate<br/>time series imputation. Observed values are projected onto<br/>low-dimensional latent space and coarse values of the missing<br/>data are reconstructed without knowing their ground truth<br/>values by this unsupervised learning approach. Finally, the<br/>reconstructed values are fed into a conditional diffusion model<br/>to obtain the precise imputed values of the time series. In<br/>this way, LSSDM not only possesses the power to identify the<br/>latent distribution but also seamlessly integrates the diffusion<br/>model to obtain the high-fidelity imputed values and assess the<br/>uncertainty of the dataset. Experimental results demonstrate<br/>that LSSDM achieves superior imputation performance while<br/>also providing a better explanation and uncertainty analysis<br/>of the imputation mechanism. The website of the code is<br/>https://github.com/gorgen2020/LSSDM imputation.<br/>Index Terms—Diffusion model, multivariate time series, imputation, variational graph autoencoder}},
  author       = {{Liang, Guojun and Abiri, Najmeh and Hashemi, Atiye Sadat and Lundström, Jens and Byttner, Stefan and Tiwari, Prayag}},
  language     = {{eng}},
  title        = {{Latent Space Score-based Diffusion Model for Probabilistic Multivariate Time Series Imputation}},
  url          = {{http://dx.doi.org/10.48550/arXiv.2409.08917}},
  doi          = {{10.48550/arXiv.2409.08917}},
  year         = {{2025}},
}